aws-samples / amazon-transcribe-post-call-analytics

Apache License 2.0
76 stars 67 forks source link

An issue with a text Duplication for AGENT and CUSTOMER. #229

Open Deuris opened 9 months ago

Deuris commented 9 months ago

Hey,

We keep experiencing an issue with some amount of the files. There is duplicated text for AGENT and CUSTOMER. Only AGENT is shown as a speaker for the converted file to MONO. Is there any solution?

Thank you in advance.

Screenshot 2024-01-23 at 15 49 36
chrislott commented 9 months ago

Hi Deuris,

Are you converting a file from mono to stereo using a tool like ffmpeg? If so, that would cause both speakers to be on both channels, which would cause this issue. Can you try uploading the original mono audio file and see what the results are?

Chris

debimishra89 commented 5 months ago

@chrislott , I am going to start a POC using Transcribe call analytics with agent-customer call recordings. Can 1-channel audio be processed correctly in Transcribe call analytics service?

chrislott commented 5 months ago

Transcribe Call Analytics only supports 2 channel audio- and you must provide which channel is which participant (customer or agent). In mono audio files, Post-Call Analytics will revert to Transcribe standard.

debimishra89 commented 5 months ago

@chrislott okay, understood. Thanks If i input 2-channel audio where both the channels are used by both the speakers (the original issue), then agent-customer text duplication happens. Is there any way to avoid this duplication?

is there any tool to preprocess the voice data to make it suitable for Transcribe call analytics? Is there any way to process the output of Transcribe Standard in Transcribe call analytics to get sentiments and out-of-box features.

Deuris commented 3 months ago

Okay, I have found the "right" 2 channeled call dialogues examples, and it works fine. BUT, what if I want to analyze calls I got from an IP telephony CRM that doesn't record calls in 2 channels? What do you I do?